stark0xbt
@stark0xbt
I track the signals operators drop before consensus catches on
AI | tech | infrastructure
DM for partnerships
123 Following 3.8K Followers
Jensen Huang spent 33 years building Nvidia. It took one day to lose it. Elon already wants him.
He was also the first person to believe in Jensen.
“Nobody in the world wanted it. I had no purchase orders, not one.”
That was 2016. Jensen had just built the DGX-1, a $300,000 AI supercomputer. Elon was his only buyer.
Jensen boxed it up himself and drove it to a small nonprofit in San Francisco. That nonprofit was OpenAI.
The math since then:
2016: $300,000 for 1 petaflop
2025: $4,000 for the same petaflop
75x cheaper in 9 years
Yesterday Nvidia’s stock opened down 13%.
Hours earlier, someone asked Elon to hire him.
His answer: “It would be an honor.”
His first customer might be his next boss.
Show more
Jensen Huang spent 33 years building Nvidia. It took one day to lose it. Elon already wants him.
He was also the first person to believe in Jensen.
“Nobody in the world wanted it. I had no purchase orders, not one.”
That was 2016. Jensen had just built the DGX-1, a $300,000 AI supercomputer. Elon was his only buyer.
Jensen boxed it up himself and drove it to a small nonprofit in San Francisco. That nonprofit was OpenAI.
The math since then:
2016: $300,000 for 1 petaflop
2025: $4,000 for the same petaflop
75x cheaper in 9 years
Yesterday Nvidia’s stock opened down 13%.
Hours earlier, someone asked Elon to hire him.
His answer: “It would be an honor.”
His first customer might be his next boss.
Show more
David Sacks says the biggest risk of AI isn’t the Terminator. It’s 1984.
“AI that lies to you, that distorts an answer, that rewrites history in real time.”
His point: AI is becoming the main way people get information online. Whoever controls the model controls what millions of people believe is true.
That’s the whole reason Elon built Grok around one goal: maximally truth-seeking.
Every lab says its model is safe. Very few will say who decides what “safe” means.
“It’s not the Terminator, it’s 1984.”
Show more
David Sacks says the biggest risk of AI isn’t the Terminator. It’s 1984.
“AI that lies to you, that distorts an answer, that rewrites history in real time.”
His point: AI is becoming the main way people get information online. Whoever controls the model controls what millions of people believe is true.
That’s the whole reason Elon built Grok around one goal: maximally truth-seeking.
Every lab says its model is safe. Very few will say who decides what “safe” means.
“It’s not the Terminator, it’s 1984.”
Show more
Gavin Baker sat with TBPN for 33 minutes and gave real numbers on SpaceX. I ran the math on all of them.
A gigawatt of AI compute on Earth costs $60B. Baker puts the same gigawatt in orbit at about $30B. “You don’t need that in space.”
That is half the cost of Earth.
Here’s the math. $25B of every Earth gigawatt is power and cooling. Reusable Starship launches it for about $5B. Even on my numbers that’s $40B in orbit.
If that holds, every data center going up on Earth right now is the expensive version. And the only company that can launch the cheap one is SpaceX.
The revenue side is the part nobody is modeling. The Google deal pays SpaceX $50B per gigawatt, 2–3x what an average neocloud gets. 4 GW is $200B.
SpaceX trades at about $1.8T. The market is pricing rockets and Starlink. Compute in orbit isn’t in the model yet.
The tell for me: 10,000+ employees bought on the IPO. They had a chance to sell every six months for 10 years. The people with the best information bought.
All 8 numbers with the formulas are in the video. Save it.
Show more
Gavin Baker owned 15% of Nvidia and 10% of Tesla sub-$2 billion. His biggest regret won't be either. It'll be the price you'll pay for SpaceX after the IPO.
"A company like this comes once in your career. Thank God I took it. Plenty of people didn't."
His analog: British East India Company. Empire scale, not tech scale.
The setup:
- 10,000 SpaceX employees could have sold every 6 months for a decade
- Almost nobody did
- Secondary pricing is a decade underpriced
Baker on the Elon-Anthropic partnership: "Four months ago it didn't exist. Then it made sense. Everyone has a price."
Incumbents realign around SpaceX gravity when it moves.
The IPO doesn't price the company. It prices the last window.
Show more
Elon just connected his two biggest bets in one sentence.
Starlink’s software was built to move enormous amounts of data through space. Turns out it’s also very good at training superintelligence.
That’s not a metaphor. This week the Grok team dumped JAX and SGLang and rewrote the entire RL and inference stack from scratch in C/C++. Elon says it’s built on the work of the Starlink engineers.
The same people who keep 10,000 satellites in sync down to the leap second now write Grok’s training stack. And the network is heading to 100,000.
SpaceX trades at about $1.8T, priced mostly as rockets and Starlink. Elon just told you the Starlink team is building the AI too.
All 8 numbers with the formulas are in the video. Save it.
Show more
Turns out that the Starlink software, built for managing the movement of vast amounts of data through space and to the ground, is also very well suited to training super intelligence
Elon just connected his two biggest bets in one sentence.
Starlink’s software was built to move enormous amounts of data through space. Turns out it’s also very good at training superintelligence.
That’s not a metaphor. This week the Grok team dumped JAX and SGLang and rewrote the entire RL and inference stack from scratch in C/C++. Elon says it’s built on the work of the Starlink engineers.
The same people who keep 10,000 satellites in sync down to the leap second now write Grok’s training stack. And the network is heading to 100,000.
SpaceX trades at about $1.8T, priced mostly as rockets and Starlink. Elon just told you the Starlink team is building the AI too.
All 8 numbers with the formulas are in the video. Save it.
Show more
Turns out that the Starlink software, built for managing the movement of vast amounts of data through space and to the ground, is also very well suited to training super intelligence
Elon just said Grok 4.8 will have “nothing in common with 4.7 (thankfully).” It soft-launches for Grok Bot users this week.
So I ran the math on the version he’s replacing.
Grok 4.7 charges $6 per 1M output tokens. Claude Opus 5.5 charges $20. OpenAI’s GPT-6 Astra charges $50.
That is 3.3x cheaper than Opus and 8.3x cheaper than Astra.
It isn’t the smartest model. Opus scores 58 on the Intelligence Index, Grok 4.7 scores 35. But divide list price by score and Grok still wins. $0.17 per point. Opus is $0.34. Astra is $0.88.
Then demand broke it. Daily users went from under 100K to over 600K in about eight weeks. About 800K pay every week.
On Nov 1 Elon said serving it was “unsustainably high relative to revenue.” Limits got cut. On Nov 9, after a roundtable with the engineering team, limits were reset.
Now the whole RL and inference stack has been rewritten from scratch in C/C++. Pre-training is done. 4.8 is in RL and evals are “starting to climb.”
The cheapest model on the list is the one Elon is glad to leave behind. That’s the setup.
All 8 numbers with the formulas are in the video. Save it.
Show more
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.
Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes.
Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
Show more
Elon just said Grok 4.8 will have “nothing in common with 4.7 (thankfully).” It soft-launches for Grok Bot users this week.
So I ran the math on the version he’s replacing.
Grok 4.7 charges $6 per 1M output tokens. Claude Opus 5.5 charges $20. OpenAI’s GPT-6 Astra charges $50.
That is 3.3x cheaper than Opus and 8.3x cheaper than Astra.
It isn’t the smartest model. Opus scores 58 on the Intelligence Index, Grok 4.7 scores 35. But divide list price by score and Grok still wins. $0.17 per point. Opus is $0.34. Astra is $0.88.
Then demand broke it. Daily users went from under 100K to over 600K in about eight weeks. About 800K pay every week.
On Nov 1 Elon said serving it was “unsustainably high relative to revenue.” Limits got cut. On Nov 9, after a roundtable with the engineering team, limits were reset.
Now the whole RL and inference stack has been rewritten from scratch in C/C++. Pre-training is done. 4.8 is in RL and evals are “starting to climb.”
The cheapest model on the list is the one Elon is glad to leave behind. That’s the setup.
All 8 numbers with the formulas are in the video. Save it.
Show more
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.
Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes.
Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
Show more
Gavin Baker says no big tech winner ever won by being the cheapest. Apple, Nvidia. Not once.
AI is the first time that flips.
His frame: “the low-cost producer always wins. And we haven’t had that in tech.”
Intelligence now comes down to two numbers. Tokens per answer and cost per token. The labs own the first. Everyone else is fighting over the second.
The math he keeps quoting:
Take a GPU from 40% utilization to 80%
Cost per token drops in half
Margins jump with it
And power is capped. “Seven states have bills to make it illegal to build data centers.” So squeezing the factories you already have is the only lever left.
Then how he caught Tesla and Nvidia early: “when a technological discontinuity encounters first principles thinking.”
Elon put the battery at the bottom of the car. Jensen saw that graphics math was deep learning math.
His read: nobody has done that for AI factories yet.
Cheapest intelligence wins. First time in tech.
Show more
Elon sat with Ron Baron for 61 minutes and gave real numbers. I ran the math on all of them.
SpaceX did $18.7B last year. Elon says it exits December at a $100B run-rate. His best guess for December 2027 is $500B. “This is not a promise.”
That is 26.7x in two years.
The data center trade is the part nobody is modeling. Build for $35B in about 8 months, rent it out for about $50B a year. That pays back in 8.4 months.
Starship takes a kilo to orbit from $18,500 to under $100. That’s 185x cheaper.
Then 100 GW of compute in orbit every year. At $10B per GW, that is a new trillion in revenue every single year.
SpaceX trades at $1.82T. That’s 3.6x Elon’s 2027 run-rate.
All 8 numbers with the formulas are in the video. Save it.
Show more
Here’s the full interview of @elonmusk with Ron Baron
0:13 - Exciting times ahead
0:21 - First principles and TerraFab
2:18 - Chip fabs geopolitical risk
5:51 - Terawatt scale ambition
6:14 - Trillion dollar phased build
7:36 - Why others don’t improve
7:49 - Austin research fab iteration
10:49 - Culture of speed and ideas
13:48 - Walking the floor hands-on
15:39 - Data centers pitchforks
16:44 - OpenAI origins Anthropic
18:09 - Building data centers fast
20:01 - Revenue exploding Jose meme
23:16 - Model Y best in China
23:28 - China ownership exception
24:50 - ARR to five hundred billion
27:45 - Digital AI economy boost
28:21 - Robotics five times productivity
30:10 - Rent robots not sell
30:41 - Memphis community benefits
32:55 - Data centers in space
33:35 - StarMind and space power
36:26 - Cost to orbit history
36:52 - Falcon nine reusability limits
39:09 - Hourly launches Starship size
40:25 - Chopsticks and heat shield
44:30 - Starship unlocks Starlink
46:49 - Agentic traffic thousand times
48:09 - Starlink V3 bandwidth explosion
51:26 - Spectrum and telcos
52:43 - Satellite replacement cycle
53:17 - Laser links hundred times
53:55 - Competing with China power
55:37 - Space intelligence advantage
57:19 - Demand from superintelligence
58:02 - Universal high income future
1:00:43 - Legacy contribution
Show more
Elon sat with Ron Baron for 61 minutes and gave real numbers. I ran the math on all of them.
SpaceX did $18.7B last year. Elon says it exits December at a $100B run-rate. His best guess for December 2027 is $500B. “This is not a promise.”
That is 26.7x in two years.
The data center trade is the part nobody is modeling. Build for $35B in about 8 months, rent it out for about $50B a year. That pays back in 8.4 months.
Starship takes a kilo to orbit from $18,500 to under $100. That’s 185x cheaper.
Then 100 GW of compute in orbit every year. At $10B per GW, that is a new trillion in revenue every single year.
SpaceX trades at $1.82T. That’s 3.6x Elon’s 2027 run-rate.
All 8 numbers with the formulas are in the video. Save it.
Show more
Here’s the full interview of @elonmusk with Ron Baron
0:13 - Exciting times ahead
0:21 - First principles and TerraFab
2:18 - Chip fabs geopolitical risk
5:51 - Terawatt scale ambition
6:14 - Trillion dollar phased build
7:36 - Why others don’t improve
7:49 - Austin research fab iteration
10:49 - Culture of speed and ideas
13:48 - Walking the floor hands-on
15:39 - Data centers pitchforks
16:44 - OpenAI origins Anthropic
18:09 - Building data centers fast
20:01 - Revenue exploding Jose meme
23:16 - Model Y best in China
23:28 - China ownership exception
24:50 - ARR to five hundred billion
27:45 - Digital AI economy boost
28:21 - Robotics five times productivity
30:10 - Rent robots not sell
30:41 - Memphis community benefits
32:55 - Data centers in space
33:35 - StarMind and space power
36:26 - Cost to orbit history
36:52 - Falcon nine reusability limits
39:09 - Hourly launches Starship size
40:25 - Chopsticks and heat shield
44:30 - Starship unlocks Starlink
46:49 - Agentic traffic thousand times
48:09 - Starlink V3 bandwidth explosion
51:26 - Spectrum and telcos
52:43 - Satellite replacement cycle
53:17 - Laser links hundred times
53:55 - Competing with China power
55:37 - Space intelligence advantage
57:19 - Demand from superintelligence
58:02 - Universal high income future
1:00:43 - Legacy contribution
Show more
Ron Baron told Elon that Starlink alone could be worth $15 trillion. All of SpaceX trades under $2 trillion.
Then he admitted even that is probably too low.
His math: Starlink doing “a trillion dollars in revenues,” “seven or eight hundred billion dollars in EBITDA in 10 years,” and “worth 14 or 15 trillion.”
Today the whole company, rockets and xAI included, is worth less than two.
His frame: “Now I think we’re probably really way low on how much Starlink can be worth.” Because now SpaceX has space data centers too.
Elon didn’t push back. He added:
“Starlink revenue is growing very rapidly. It’s like you’re roughly doubling revenues per year at large scale.”
Then the line that matters: “AI does exceed that. With AI you have order of magnitude per year growth potential.”
Starlink at $15T is the conservative case.
Show more
Everyone else takes 2.5 years and $60B to build a data center. Elon does it in 8 months for $35B.
Ron Baron laid out the math to Elon’s face at his conference in New York.
The industry: $60B to build one gigawatt. Five-year leases. Maybe $12B a year in rent.
SpaceX: $35B. Eight or nine months. Then it rents that compute for around $50B a year.
$50B a year on a $35B build. The rest of the industry needs five years to earn back less.
Then Elon added the number nobody expected:
“Q2 to Q3 revenue increase was, I think, over 80 percent.”
And next: “Probably something like 70 percent, Q3 to Q4.”
His frame: “It is extraordinarily difficult to bring on compute at scale.”
Everyone else is still waiting on power. SpaceX is already collecting rent.
Show more
Everyone else takes 2.5 years and $60B to build a data center. Elon does it in 8 months for $35B.
Ron Baron laid out the math to Elon’s face at his conference in New York.
The industry: $60B to build one gigawatt. Five-year leases. Maybe $12B a year in rent.
SpaceX: $35B. Eight or nine months. Then it rents that compute for around $50B a year.
$50B a year on a $35B build. The rest of the industry needs five years to earn back less.
Then Elon added the number nobody expected:
“Q2 to Q3 revenue increase was, I think, over 80 percent.”
And next: “Probably something like 70 percent, Q3 to Q4.”
His frame: “It is extraordinarily difficult to bring on compute at scale.”
Everyone else is still waiting on power. SpaceX is already collecting rent.
Show more
Ron Baron turned $100M into $70B. His own math: $30B of that is Elon Musk.
Read that again. One founder is worth $30 billion to a single fund.
Without Elon, Baron is still a legend. 19% a year for 34 years. Almost double the market.
With Elon, it’s 21%. Two points a year. Over three decades, that’s the difference between $40B and $70B.
That’s what compounding on the right person looks like.
Most fund managers spend their careers diversifying away from exactly this kind of bet. Baron did the opposite. He kept buying, kept holding, and never sold the first chance he got.
His frame: “He’s probably the only person in this country who is able to build things at scale at attractive prices. Only one.”
Wall Street calls it concentration risk.
Baron calls him “a once in a lifetime person.”
$30 billion says Baron was right.
Show more
Elon has been saying this for years. It’s the whole reason behind Mars. A civilization that stays on one planet is betting everything on one life cycle.
Elon Musk on why no civilization lasts forever:
“You do tend to see the same patterns, similar patterns for civilizations, where they go through a life cycle like an organism.”
“Just like a human: zygote, fetus, baby, toddler, teenager… eventually gets old and dies.”
“The civilizations go through a life cycle. No civilization will last forever.”
The question is where we are in ours.
Show more
Gavin Baker owned 15% of Nvidia and 10% of Tesla before either was worth $2 billion. He’s 50. And he just admitted Claude Code was too hard for him, so he switched to Grok Bot.
“I glued together all this stuff with Claude Code and, you know, I’m sure it was user error on my part. 50 years old. I’m doing my best.”
Then he tried Grok Bot.
“Grok Bot is the most impactful AI product to me and to the firm that we’ve had yet.”
“It just makes it so easy. I think I’m up to like 17 Grok Bots, and every day I think of a new one to add.”
And it’s not a toy for him. It’s how he runs money:
“For the more professional use cases, Grok Bot has just been transformative for me.”
“So much of investing is making sure that the right information gets to the right person at the right time, and Grok Bot is so good at that.”
One of the best tech investors alive runs his firm on 17 AI agents. And he’s adding a new one every day.
Show more
Gavin Baker owned 15% of Nvidia and 10% of Tesla when both were sub-$2 billion. He still says SpaceX is different.
“It’s possible that you have an opportunity to invest in a company like that one time in your career. And thank god you did it, because there’s plenty of people that had the opportunity and didn’t.”
His analog: if SpaceX puts a city on the Moon and then Mars, it “becomes the British East India Company of the solar system.”
The detail nobody modeled: employees “had a chance to sell every six months for the last 10 years.” Instead, “over 10,000 SpaceX employees bought on the IPO.”
On Elon and Anthropic: “Everyone has a price. You look at Elon and Anthropic’s relationship four months ago… as the opportunity unfolded, it made sense for them to partner.”
15% of Nvidia. 10% of Tesla. And SpaceX is still the one he calls once-in-a-career.
Show more
Ron Baron says SpaceX hits $50 trillion. Elon told him that’s too low.
“I think it’ll be worth 20 trillion, 30 trillion, 50 trillion. Elon thinks I’m way low.”
It’s at $2 trillion today.
“This is not pulling numbers out of the air.”
Starlink alone: “Somewhere around 14 trillion dollars, 14, 15 trillion dollars.”
Then data centers in space: “That’s unbelievable. You can’t build them on the planet anymore.” He expects that business to be a multiple of Starlink.
His target: “We’re going to make at least 20 or 30 times.”
56 years of investing. And Elon still says he’s too conservative.
Show more
